biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core] Pipeline Parallel Support (#4412)

Browse Source 
Signed-off-by: Muralidhar Andoorveedu <muralidhar.andoorveedu@centml.ai>
This commit is contained in:
Murali Andoorveedu
2024-07-02 10:58:08 -07:00
committed by GitHub
parent 15aba081f3
commit c5832d2ae9
82 changed files with 1096 additions and 400 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/executor/multiproc_gpu_executor.py
View File

@@ -91,17 +91,17 @@ class MultiprocessingGPUExecutor(DistributedGPUExecutor):
self,
method: str,
*args,
async_run_remote_workers_only: bool = False,
async_run_tensor_parallel_workers_only: bool = False,
max_concurrent_workers: Optional[int] = None,
**kwargs,
) -> Any:
"""Runs the given method on all workers.
Args:
async_run_remote_workers_only: If True the method will be run only
in the remote workers, not the driver worker. It will also be
run asynchronously and return a list of futures rather than
blocking on the results.
async_run_tensor_parallel_workers_only: If True the method will be
run only in the remote TP workers, not the driver worker.
It will also be run asynchronously and return a list of futures
rather than blocking on the results.
"""
if max_concurrent_workers:
@@ -114,7 +114,7 @@ class MultiprocessingGPUExecutor(DistributedGPUExecutor):
for worker in self.workers
]
if async_run_remote_workers_only:
if async_run_tensor_parallel_workers_only:
# Just return futures
return worker_outputs